Digital Humanities and Data
Francesca Giannetti
2022-10-19
Abstract
Is there a humanist way of working with data (cf. Posner, “Data Trouble”)? This workshop prods at that question, examining tensions with quantification and categorization, and moving towards (we hope) repair and remediation. A hands-on portion will look at concrete ways to represent research topics as tabular data (aka a spreadsheet), one of the handiest and most portable of data formats.
Introduction
This workshop introduces the challenges and benefits of translating humanities sources into machine-accessible formats. It also provides a gentle introduction to programming in R using the RStudio environment.
Take a look around: you’re in RStudio, an environment that makes coding in R more transparent. Your window is divided into four segments. In the upper right, you’ll see the environment: this displays all the objects you have loaded into memory (currently none). In the upper left, you’ll see the script editor: this is the place to work on code that you’re currently writing (or borrowing from elsewhere!). To run code in code chunks (the grey chunks), you can either press Ctrl+Enter to run single lines or click the green arrow to run the entire chunk. In the lower left, you’ll see the console: this is where code actually executes and where the output prints. In the lower right, you’ll see a few different things. The “Files” tab shows whatever files live in the directory (or folder) that R is currently working in; if you run any plots, they’ll show up in the “Plots” tab; you can also get help in the “Help” tab.
To start, if you’re using your own computer, run the section of setup code above by clicking the green arrow in the upper right of the grey box.
Humanities Data Tensions
In her talk, “Data Trouble,” Miriam Posner enumerates some of the reasons behind the data negativity in some humanities circles. It goes beyond a fondness for close reading and an aversion to positivism. Those reasons fall under the sign of categorization. E.g.:
- demarcation
- parameterization
- ontological stability
- replicability
- boundedness
- deracination
As an example of the first, Posner discusses the culturomics authors’ paper, specifically the visualization on inventions and their adoption rate.
A figure from “Quantitative Analysis of Culture…”
Demarcation
Demarcation: the separation of research observations into discrete items in order to perform operations on them.
In the figure above, the authors see proof that more recent cohorts of inventions were adopted more rapidly than previous cohorts. The visualization suggests that each invention cohort appeared like a bolt out of the blue. Humanists, on the other hand, tend to see inventions emerging from constellations of forces, integrating previous technologies in the succeeding technology. Not separate, in other words. And yet, separation is a necessary precondition for data manipulation and analysis. The computer can’t do much with an undifferentiated mass of information.
Parameterization
Parameterization: a standard scale of measurement to facilitate comparisons.
Some things lend themselves well to parameterization: word counts, canvas size, the length of a song, or the shape of a pot. But within cultural artifacts, many features that are legible to human perception are difficult to parameterize, e.g. tone of voice, posture, facial expression. And some things that seem simple to measure, e.g. how far it is from London to Edinburgh, change considerably depending on the historical period, mode of transport, class status, gender, weather, and so on.
Ontological Stability
Ontological stability: how we organize entities to make a data model.
Ontologies get at the way in which one sees the world. To humanists, perspective changes things, occasionally in significant ways. Take for example this runaway slave advertisement from 1752 and its current (2022) representation in the New Jersey Slavery Records.
A runaway slave ad compared to its representation in NJ Slavery
Replicability
Replicability: the benefit of published data is reuse and replicability, but cui bono in the humanities?
As Posner suggests, the interpretation of the data IS the work in the humanities, and it would be foreign to most humanists to suggest that any other scholar reach the same conclusions as they do. With that said, publishing well-structured and well described humanities datasets does facilitate building on previous data-centric work. That does count for a lot.
Boundedness
Boundedness: all historical records are partial, therefore no historical dataset is representative?
We have a fairly limited vocabulary for referencing gaps or absences in our data, but there are some strategies.
Deracination
Deracination: creating data involves capturing some aspects of our sources and disregarding others. In other words, some context is inevitably lost.
It’s not possible to represent every feature of a given source. But fortunately, that is okay. No humanist would assert that a digitized book is the same as the print book used as its source. They both accomplish different purposes, and have different affordances and drawbacks.
To take another example, from the correspondence edition of the War Service Bureau Records, the following figure shows the editorial pipeline from manuscript letter, to TEI XML encoding of the same letter, to its HTML representation on the website. The web edition makes no effort to capture many of the material aspects of the source. It focuses on content and editorial annotation instead, on the assumption that readers can always refer back to the digitized or print manuscript.
A letter from the Rutgers College War Service Bureau Records and its representations as TEI-XML and HTML
Data Modeling
A few words on data modeling in the humanities… or in any discipline, really.
This is a translation of Herbert Stachowiak’s Allgemeine Modelltheorie (General Model Theory) published in 1973, and borrowed from modelpractice.
Fundamental Model Properties
- Mapping: Models are always models of something, i.e. mappings from, representations of natural or artificial originals, that can be models themselves.
- Reduction: Models in general capture not all attributes of the original represented by them, but rather only those seeming relevant to their model creators and/or model users.
- Pragmatism: Models are not uniquely assigned to their originals per se. They fulfill their replacement function a) for particular – cognitive and/or acting, model using subjects, b) within particular time intervals and c) restricted to particular mental or actual operations.
In other words, a lot of the tensions identified by Posner are good and normal insofar as we’re aware of our individual contexts as researchers, and we document what we’re trying to represent, and what we’ve determined to be out of scope.
Data from Archival Sources
Some archival sources can be readily understood as data. Take, for example, this antebellum American newspaper’s subscription book. This is already a technology for organizing and storing data; we simply want to transcribe it so that we can represent and analyze it more effectively. Depending on one’s research, one may want to quantify subscribers by gender, map their locations, or calculate turnover.
